DEL-Ranking: Ranking-Correction Denoising Framework for Elucidating Molecular Affinities in DNA-Encoded Libraries

ICLR 2026 Conference Submission19969 Authors

Keywords: DNA Encoded Library, Deep learning, Drug design

Abstract:

DNA-encoded library (DEL) screening has revolutionized protein--ligand binding detection by enabling efficient exploration of vast chemical spaces through read count analysis. Despite its transformative potential, two critical challenges limit its effectiveness: (1) stochastic noise in low copy number regimes, where Poisson fluctuations significantly distort binding signals, and (2) systematic biases between observed read counts and actual binding affinities due to experimental artifacts and amplification variability.
We introduce DEL-Ranking, a comprehensive framework that addresses these dual challenges through targeted innovations. To mitigate stochastic noise, we incorporate a dual-perspective ranking mechanism that prioritizes stable relative ordering relationships over volatile absolute counts. To bridge the read count-affinity gap, our Chemical-Referenced Correction (CRC) module identifies critical binding-related functional groups and leverages these structure-activity insights to guide precise count adjustments.
A key contribution is our release of three novel DEL datasets featuring 2D molecular sequences, 3D conformational data, and functionally-derived activity labels—addressing a significant resource gap in the field and enabling more robust method development.
Rigorous validation across multiple datasets reveals that DEL-Ranking consistently outperforms existing methods, achieving a remarkable 28\% improvement in Spearman correlation even under high-noise conditions. Our framework both enhances identification of high-affinity compounds and reveals novel functional motifs--Pyrimidine Sulfonamide, beyond known Benzene Sulfonamide groups. These interpretable insights accelerate therapeutic candidate discovery while advancing understanding of molecular recognition mechanisms.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 19969